A Varying Coefficient Model to Jointly Test Genetic and Gene-Environment Interaction Effects

Most human traits are influenced by the interplay between genetic and environmental factors. Many statistical methods have been proposed to screen for gene-environment interaction (GxE) in the post genome-wide association study era. However, most of the existing methods assume a linear interaction between genetic and environmental factors toward phenotypic variations, which diminishes statistical power in the case of nonlinear GxE. In this paper, we present a flexible statistical procedure to detect GxE regardless of whether the underlying relationship is linear or not. By modeling the joint genetic and GxE effects as a varying-coefficient function of the environmental factor, the proposed model is able to capture dynamic trajectories of GxE. We employ a likelihood ratio test with a fast Monte Carlo algorithm for hypothesis testing. Simulations were conducted to evaluate validity and power of the proposed model in various settings. Real data analysis was performed to illustrate its power, in particular, in the case of nonlinear GxE.

Automated summary

This study presents a flexible statistical procedure to detect gene-environment interaction (GxE) regardless of whether the underlying relationship is linear or not. By modeling the joint genetic and GxE effects as a varying-coefficient function of the environmental factor, the proposed model is able to capture dynamic trajectories of GxE. A likelihood ratio test with a fast Monte Carlo algorithm is employed for hypothesis testing. Simulations and real data analysis demonstrate the validity and power of the proposed model in various settings, especially for nonlinear GxE.